Edge-preserving Multi-scale Image Decomposition And Its Applications

For many applications in digital image processing, we often need to operate on images at multiple scales. Particularly, in computational photography, images are often decomposed into a piecewise smooth base layer and one or more detail layers. Frequently, edge-preserving image smoothing is the crux to such image multi-scale decomposition.In this paper we introduce two new edge-preserving image smoothing algorithms. The first one is the similar pixel based edge-preserving image smoothing algorithm. For every pixel in an image, we set new intensity value for the pixel as the average value of those similar colors in its neighborhood. If there is no similar color in its neighbor, this pixel will be deal as noise. The second algorithm for edge-preserving image smoothing is by compression of residuals obtained by some known smoothing methods. It is observed that when the edge of the smooth image is blurred, the corresponding residual amplitude will be larger. The gradient of the edge in the image will be larger as well, especially, in salient edges. Based on these two facts, a smooth image is first acquired with possibly blurred edges by some common smoothing method, such as Gaussian filtering, median filtering. The difference image between the original image and the smooth one is defined as the residual. Second, the amplitudes of the difference image are compressed nonlinearly, by which large magnitudes have been decreased much while small details are preserved well. Finally, an edge preserving smooth image-is obtained by subtracting the compressed difference image from the original one. By the residual compression scheme, any existing image smoothing filter can be used for edge-preserving image decomposition purpose.The appealing characteristics of our proposed algorithms are simple, easy to implement, linear time complexity and low memory spending. We have applied the two edge-preserving image smoothing algorithms for multi-scale image decomposition as well as many applications of image decomposition, for example, detail enhancement, image abstraction, HDR Image Compression, image tone tuning, etc. We also extend these image smoothing and decomposition algorithms to videos. From the examples we have experimented, we find that our proposed algorithms can achieve similar or even higher quality results as compared with other known image decomposition algorithms. Even though, our algorithms own a predominant benefit of high speed than other methods.